Study On Key Technologies Of Peer-to-Peer Computing Systems Based On Theories Of Complex Networks

Due to extreme dynamics and complexity, peer-to-peer (P2P) computing systems can beviewed as representative Complex Networks, and some of their topologies have been provedto exhibit strong power-law degree distributions and small-world phenomena found in manyother complex networks. However, the main difficulty in designing such systems is that cur-rently, very little is known about the nature of topologies on which they would be operating.The end result is that even simple algorithms or protocols, as in the case of Gnutella, couldresult in complex interactions that can adversely affect system performance.In this thesis, we eye our viewpoint upon these topological properties and conduct anextensive study on key technologies of P2P networks, by managing to answer the followingquestions one by one:how to characterize these topological properties found in real-world P2P networks, andthen how to generate topologies that truly possess such properties?how to better utilize these properties to improve the performance of P2P networks? andhow to further utilize these properties to enhance attack-survivability of P2P networksin case of virus spreading and malicious attacks?The main contributions of this thesis are:To answer the first question, we propose FitGnu, a novel topology generator, by in-troducing techniques like (1) explicitly self-sustaining power-law degree distributions duringtopology evolution, (2) fitness-based preferential attachment, (3) utilization of semantic clus-tering and degree-degree correlation, (4) consideration of both local linking and global link-ing based on geographical proximity between nodes, and (5) allowing nodes'joining/leavingthe network dynamically. The FitGnu topology generator has been verified, theoreticallyand experimentally, to be able to generate topologies that can give a more comprehensiveand precise characterization of topologies of real-world Gnutella-like P2P networks. As a response to the second question, we propose ToA3, a new scalable P2P file-sharingsystem to tackle serious scaling problems of current unstructured P2P networks. The key ideaof ToA3 is to produce an overlay topology with desirable properties, adapt peers towardsbetter neighbors and direct queries to right next hops, with as few duplicated messagesas possible. To achieve this goal, several innovative techniques are introduced into ToA3:(1) DYTOPA ? a dynamic topology adaptation algorithm, (2) SMARTSEARCH ? a high-performance search algorithm specially-designed for DYTOPA, (3) HOTREPLICATION ? ahot-data replication strategy designed for better load balancing, and (4) BIGDOWNLOAD ?an efficient QoS-provisioning strategy for large-sized file download in P2P environments.We deploy and compare ToA3 with the other three existing P2P systems over both static anddynamic environments, and the results prove significant performance gains of ToA3.To answer the last question, we propose DeepCure, a novel targeted but decentralizedimmunization strategy, to defense intentional attacks in scale-free P2P networks. Differentfrom existing strategies, DeepCure identifies immunization targets as not only the mosthighly-connected nodes but also the nodes with highest availability and/or highest link load,with the aim of injecting immunization information into just right targets to cure. To bettertrade off the cost and the efficiency, DeepCure deliberately select these targets from 2-localneighborhood, as well as topologically-remote but semantically-close friends if needed. Toremedy the weakness of existing strategies in case of sudden epidemic outbreak, DeepCureis also coupled with LHRT, a local-hub oriented rate throttling mechanism, to enforce proac-tive rate control. Extensive simulation results show that DeepCure outperforms its competi-tors, producing an arresting increase of the network attack tolerance, at a lower price ofeliminating viruses or malicious attacks.The work conducted in this thesis is, to our best knowledge, the first study in this areato apply the latest theories of Complex Networks to all of the above fields, and to investi-gate intrinsic matters that hinder the improvement of the functionality and performance ofGnutella-like P2P systems with a viewpoint of topological properties throughout the thesis.